Multimodal fusion workflow for target delineation in cardiac radioablation of ventricular tachycardia

BackgroundCardiac radioablation (CR) is an innovative treatment to ablate cardiac arrythmia sources by radiation therapy. CR target delineation is a challenging task requiring the exploitation of highly different imaging modalities, including cardiac electro-anatomical mapping (EAM). PurposeIn this work, a data integration process is proposed to alleviate the tediousness of CR target delineation by generating a fused representation of the heart, including all the information of interest resulting from the analysis and registration of electro-anatomical data, PET scan and planning computed tomography (CT) scan. The proposed process was evaluated by cardiologists during delineation trials. MethodsThe data processing pipeline was composed of the following steps. The cardiac structures of interest were segmented from cardiac CT scans using a deep learning method. The EAM data was registered to the cardiac CT scan using a point cloud based registration method. The PET scan was registered using rigid image registration. The EAM and PET information, as well as the myocardium thickness, were projected on the surface of the 3D mesh of the left ventricle. The target was identified by delineating a path on this surface that was further projected to the thickness of the myocardium to create the target volume. This process was evaluated by comparison with a standard slice-by-slice delineation with mental EAM registration. Four cardiologists delineated targets for three patients using both methods. The variability of target volumes, and the ease of use of the proposed method, were evaluated. ResultsAll cardiologists reported being more confident and efficient using the proposed method. The inter-clinician variability in delineated target volume was systematically lower with the proposed method (average dice score of 0.62 vs. 0.32 with a classical method). Delineation times were also improved. ConclusionsA data integration process was proposed and evaluated to fuse images of interest for CR target delineation. It effectively reduces the tediousness of CR target delineation, while improving inter-clinician agreement on target volumes. This study is still to be confirmed by including more clinicians and patient data to the experiments.